This invention relates to a charge-coupled device for spectroscopic detection.
In emission spectroscopy, a specimen is heated to incandescence and the light emitted by the specimen is collimated, dispersed in accordance with wavelength, and focused in an image plane. The intensity of light in the image plane varies along a wavelength axis, and FIG. 1 illustrates variation of intensity in the image plane with position along the wavelength axis. In absorption spectroscopy, a beam of light is incident on an absorbent specimen, and the transmitted light is dispersed in accordance with wavelength and is focused in an image plane. In each case, the distribution of light intensity as a function of wavelength is dependent on the composition of the specimen.
It is known to measure the distribution of light intensity as a function of wavelength in emission or absorption spectroscopy by use of a charge-coupled device (CCD). FIG. 2 illustrates schematically a known form of CCD 10. The CCD shown in FIG. 2 comprises a die made of monocrystalline silicon into which various impurities have been implanted using conventional integrated circuit fabrication techniques. The pattern of impurity type and concentration defines a rectangular sensing region 14 composed of multiple columns extending perpendicular to the longer dimension of the sensing region 14. For convenience, it will be assumed in the following discussion that wavelength varies along a horizontal axis and that the columns are vertical, as shown in FIG. 2.
A three-phase frame electrode structure 16 (FIG. 3) traverses the columns of the sensing region. The frame electrode structure is connected to a three-phase clock driver (not shown) that applies selected voltage levels to the electrodes and thereby establishes a potential profile as shown by the dashed line 19, dividing each column into multiple pixels 18. It will be understood that although the sensing region 14 shown in FIG. 2 is composed of only 100 pixels, a practical CCD for spectroscopic detection might have well over 100,000 pixels. Also, although, the pixels are shown in FIG. 2 as being square, this is not necessary.
The CCD is placed in the image place of the spectrometer so that the longer dimension of the rectangular sensing region is disposed parallel to the wavelength axis of the spectrum provided by the spectrometer. Therefore, each column of pixels is associated with an interval in the wavelength range. Photons that are incident on a particular pixel of the sensing region result in generation of photoelectrons in the semiconductor die at a rate that depends on the intensity of light incident on the pixel, and photoelectrons that do not recombine with holes are retained in the pixel by potential barriers that bound the pixel. Thus, the size of the charge packet accumulated in a given column of pixels during an exposure interval is representative of the intensity of light within the wavelength range associated with that column, and the distribution of size of charge packets along the wavelength axis represents the distribution of light intensity as a function of wavelength in the light beam provided by the sample.
Along one horizontal edge of the rectangular sensing region there is a readout register 22. As shown in FIG. 2, the readout register extends parallel to the wavelength axis. Referring to FIG. 3, the readout register 22 is bounded along its edge that is farther from the sensing region 14 by a potential barrier 28. The readout register comprises one charge transfer cell 26 for each column of pixels in the sensing region of the CCD. The readout register also comprises a so-called floating diffusion 30 at one end. The floating diffusion is coupled to a readout amplifier 34.
The clock driver enables voltages to be applied to the frame electrode structure in a sequence that allows the charge packet accumulated in each column of pixels during the exposure period to be transferred into the corresponding transfer cell 26 of the readout register 22. In addition, a readout electrode structure 38 overlies the transfer cells of the readout register and by clocking the electrodes of the readout electrode structure in ordered sequence, the charge packets in the readout register are transferred to the floating diffusion.
A so-called last cell, controlled by a last gate or transfer gate electrode 40, is between each column of pixels in the sensing region and the associated transfer cell 26 of the readout register. However, this is conventional and is not relevant to the invention, and therefore will not be described further.
As charge packets are shifted from the readout register into the floating diffusion 30, the potential of the input terminal of the amplifier 34 varies in accordance with the size of the charge packets. In this manner, the variation along the wavelength axis in size of charge packets is converted to a time-varying voltage signal.
The output signal of the amplifier 34 is applied to processing circuitry (not shown) for extracting information from the signal. Generally, the readout amplifier 34 is fabricated on the same die as the CCD 10, but the processing circuitry is not.
As shown in FIG. 4, the readout register could extend in the direction perpendicular to the wavelength axis, with each column of pixels being clocked sequentially into the readout register and then accumulated in the floating diffusion at the end of the readout register. This is subject to disadvantage unless a shutter is used to prevent illumination of the array during readout, because charge accumulated during the readout operation is not added on a wavelength basis to the charge accumulated prior to the readout operation.
A typical silicon die that is processed to form a CCD has a thickness of about 0.1 mm and the sensing region extends to a depth of about 5-10 .mu.m below the surface at which the die is processed to form the CCD. The frame electrode structures lie over this surface, which is commonly known as the front side of the die.
If the optical signal is incident on the sensing region by way of the front side of the die, the optical signal is partially blocked by the frame electrode structure. In application of a CCD to imaging very faint optical signals, for example astronomical applications, it is known to thin the die from the back side to a thickness of about 10 .mu.m and illuminate the CCD from the back side of the die in order to avoid this problem.
The minimum noise level of an optical signal is the square root of the mean number of photons in the optical signal. Similarly, when the optical signal is converted to an electrical signal using a CCD, the noise level is at least the square root of the mean number of photoelectrons in the signal. Thus, to measure the electrical output signal of the CCD to a precision of 1 part in 1,000 requires that the signal be composed of at least 1 million photoelectrons, because the minimum noise of a 1 million electron signal is 1,000 electrons, or 1 part in 1,000 of the signal.
Another source of noise in the output signal of the CCD is the noise that is generated in the readout process itself. The level of noise that is generated in the readout process depends mainly on the readout amplifier. A readout amplifier that is designed to accommodate small charge packets adds less noise than a readout amplifier designed to accommodate large charge packets. Naturally, the size of the readout amplifier depends on the size of the charge packets that are expected in use. The charge capacity of the transfer cells of the readout register is representative of the size of the charge packets that are expected.
In the present state of the art, a compromise is made between the size of the charge packets that are provided to the readout register, as represented by the charge capacity of the readout register, and the readout noise of the readout amplifier, such that the dynamic range of the output signal cannot exceed about 200,000:1.
In CCDs that are currently used for spectroscopic detection, the charge capacity of the readout register is not significantly greater than the charge capacity of the sensing region. This imposes a limit on the number of electrons in the signal provided by the readout register, so that the signal to noise ratio cannot be increased by making multiple exposures and accumulating the charge packets in the transfer cells prior to readout.
A problem that arises in manufacture of integrated circuits in general and CCDs in particular is that of yield. If multiple devices are fabricated under nominally identical conditions and to nominally identical designs, a significant proportion of the finished devices will not function to specifications. Generally, the number of faulty devices will depend on the complexity of the design, since as complexity increases, tolerances normally decrease.